Graph Processing & Bulk Synchronous Parallel Model
Graph Processing & Bulk Synchronous Parallel Model
Graph Processing & Bulk Synchronous Parallel Model
You also want an ePaper? Increase the reach of your titles
YUMPU automatically turns print PDFs into web optimized ePapers that Google loves.
Fault Tolerance <br />
• Checkpoint: master instructs workers to save state to persistent <br />
storage (e.g. HDFS) <br />
– Vertex values <br />
– Edge values <br />
– Incoming messages <br />
• Master saves to disk aggregator values <br />
• Worker failure is detected using a heartbeat. <br />
• New worker is created using state from previous checkpoint <br />
(which could be several supersteps before current superstep) <br />
Lecture 14 : 590.02 Spring 13 <br />
24